Watershed crashes

[GoogleCodeExporter]

What steps will reproduce the problem?
1. running the attachment pipeline.
2. running on a x86_64 host with Fedora 8.
3. the dvn is tested on both a 2GB and 8GB machine.

What is the expected output? What do you see instead?
I expect to see some segmentation image. But the application crashes
directly and the Bug Reporting Tool from Gnome shows up immediately.

What version of the product are you using? On what operating system?

8.5.3085. The older version has the same problem tool.

Please provide any additional information below.

The watershed algorithm can run correctly by manual coding in python.

Original issue reported on code.google.com by zyz...@gmail.com on 27 Jun 2008 at 7:54

Attachments:

• watershed.tar.gz

[GoogleCodeExporter]

Thank you very much for your detailed report and the example.  I've tagged it 
for the
next release of DeVIDE.  If you've starred this issue, you'll be automatically 
kept
up to date on my progress in fixing it!

Original comment by cpbotha on 29 Jun 2008 at 7:37

• Changed state: Accepted
• Added labels: Milestone-Release8.9

[GoogleCodeExporter]

Original comment by cpbotha on 29 Jun 2008 at 8:08

[GoogleCodeExporter]

Original comment by cpbotha on 4 Dec 2008 at 1:04

• Added labels: Milestone-Release8.12

[GoogleCodeExporter]

Original comment by cpbotha on 23 Dec 2008 at 5:36

• Added labels: Milestone-Release9.1

[GoogleCodeExporter]

I've reproduced this behaviour on 8.5.3085 on Ubuntu 8.10 x86_64 as well as the
latest development build (3302) on the same platform.

Original comment by cpbotha on 28 Jan 2009 at 5:08

[GoogleCodeExporter]

* Crash happens inside vtkImageReslice.
* Remove watershed: no crash.
* Remove vtkImageCast and curvatureAnisotropicDiffusion: still crash
* Watershed is doing something very strange, but introspection on its output 
shows
nothing out of the ordinary.
* Writing away output as VTI file and then loading that in a clean network also 
crashes.
* I've attached a simplified network, new (9.1 and after) format.

I might have to postpone fixing this until the next release.

Original comment by cpbotha on 28 Jan 2009 at 5:53

Attachments:

• water_fixed_9.1_nodiffuse.dvn

[GoogleCodeExporter]

Curiouser and curiouser:

On Windows XP, current dev build, I can NOT reproduce the bug.

Original comment by cpbotha on 28 Jan 2009 at 8:34

[GoogleCodeExporter]

Finally fixed in r3304.

It looks like the bug was related to this posting:
http://www.itk.org/pipermail/insight-users/2005-November/015506.html

In short, lacking 128 bit unsigned long support in ITK causes this crash.  
unsigned
longs are 64 bits on x86 (hence no crash on XP).

I've fixed this now by relabeling the watershed output to signed shorts (max of 
32k
something basins).  Will change this to signed ints (2 billion odd watersheds) 
as
soon as my ITK build gets this support.

Original comment by cpbotha on 28 Jan 2009 at 9:57

• Changed state: Fixed